📄 中文摘要
AI 代理的问题往往不在于其执行的任务，而在于何时停止。作者在其公司运行五个 AI 代理，经过数百小时的运行，发现最常见的失败模式是代理不知道何时完成。缺乏明确退出条件的代理可能会不断优化已经足够好的工作，提出延迟完成的澄清问题，生成不必要的额外输出，或循环直到达到令牌限制或上下文窗口。这些问题不仅浪费资源，还可能导致效率低下。
退出条件模式：如何在正确的时刻停止你的 AI 代理
📄 English Summary
The Exit Condition Pattern: How to Stop Your AI Agent at the Right Moment
The challenges faced by AI agents often revolve around when to stop rather than what they do. The author operates five AI agents and has observed that the most common failure mode is an agent that does not recognize when it is done. Without a clear exit condition, an agent may continue to refine work that is already sufficient, ask clarifying questions that delay completion, generate unsolicited additional output, or loop until hitting a token limit or context window. These issues not only waste resources but can also lead to inefficiencies.